/******************************************************************** ** Image Component Library (ICL) ** ** ** ** Copyright (C) 2006-2013 CITEC, University of Bielefeld ** ** Neuroinformatics Group ** ** Website: www.iclcv.org and ** ** http://opensource.cit-ec.de/projects/icl ** ** ** ** File : ICLFilter/src/ICLFilter/GaborOp.h ** ** Module : ICLFilter ** ** Authors: Christof Elbrechter ** ** ** ** ** ** GNU LESSER GENERAL PUBLIC LICENSE ** ** This file may be used under the terms of the GNU Lesser General ** ** Public License version 3.0 as published by the ** ** ** ** Free Software Foundation and appearing in the file LICENSE.LGPL ** ** included in the packaging of this file. Please review the ** ** following information to ensure the license requirements will ** ** be met: http://www.gnu.org/licenses/lgpl-3.0.txt ** ** ** ** The development of this software was supported by the ** ** Excellence Cluster EXC 277 Cognitive Interaction Technology. ** ** The Excellence Cluster EXC 277 is a grant of the Deutsche ** ** Forschungsgemeinschaft (DFG) in the context of the German ** ** Excellence Initiative. ** ** ** ********************************************************************/ #pragma once #include #include #include #include #include #include namespace icl{ namespace filter{ /** \cond */ class ConvolutionOp; /** \endcond */ /// Applies Gabor filter operation on images \ingroup UNARY /** \section GEN General Information about Gabor Filters A short introduction to Gabor filters can be found at Wikipedia: A Gabor filter is a linear filter whose impulse response is defined by a harmonic function multiplied by a Gaussian function. Because of the multiplication-convolution property, the Fourier transform of a Gabor filter's impulse response is the convolution of the Fourier transform of the harmonic function and the Fourier transform of the Gaussian function. \f[ g(x,y;\lambda,\theta,\psi,\sigma,\gamma)=\exp(-\frac{x'^2+\gamma^2y'^2}{2\sigma^2})\cos(2\pi\frac{x'}{\lambda}+\psi) \f] where \f[ x' = x \cos\theta + y \sin\theta\, \f] and \f[ y' = -x \sin\theta + y \cos\theta\, \f] In this equation, \f$\lambda\f$ represents the wavelength of the cosine factor, \f$\theta\f$ represents the orientation of the normal to the parallel stripes of a Gabor function in degrees, \f$\psi\f$ is the phase offset in degrees, and \f$\gamma\f$ is the spatial aspect ratio, and specifies the ellipticity of the support of the Gabor function. Gabor filters are directly related to Gabor wavelets, since they can be designed for number of dilations and rotations. However, in general, expansion is not applied for Gabor wavelets, since this requires computation of biorthogonal wavelets, which may be very time-consuming. Therefore, usually, a filter bank consisting of Gabor filters with various scales and rotations is created. The filters are convolved with the signal, resulting in a so-called Gabor space. This process is closely related to processes in the primary visual cortex. The Gabor space is very useful in e.g., image processing applications such as iris recognition. Relations between activations for a specific spatial location are very distinctive between objects in an image. Furthermore, important activations can be extracted from the Gabor space in order to create a sparse object representation (cite http://en.wikipedia.org/wiki/Gabor_filter). \section THECLASS The GaborOp class The GaborOp class provides basic functionalities for applying Gabor filters on Images. To achieve optimal performance, it wraps the ConvolutionOp class to realize the internal image convolution operations. Determined by a set op input parameters, it internally creates a filter bank that caches all gabor masks. In contrast to other filters, it knows two modes: -# whole image mode in this mode, the filter bank image applied on the whole input image, and an output image is created with one channel for each filter. This mode works essentially like all other UnaryOps. -# specified center mode here, the filters are applied not on all image locations, but on some well defined image locations. The result is not an image, but a matrix of filter responses, where the matrix's x index references the convolution center, and the y index then defines the filter index on this location. (... some more detail here!) The GaborOp class provides functionalities for the creation of Gabor-Filter kernels, as well as for applying gabor filter operation on images. As mentioned above, in contrast to other convolution operations, Gabor filters are often applied as so called Gabor-Jets at some specified image locations only. A Gabor-Jet complies a stack of gabor kernels that are created by some methodical variation of one, some or all gabor mask parameters.\n Each GaborOp object provide function to create a gabor jet internally, whereas in the easiest case, there is only one value for each parameter, an consequently, only a single gabor mask is created.\n In addition to the parameters mentioned in the formula above, the size of the created gabor kernels must be set, and the parameter values must be adapted to to it. In the following, each parameter is explained again, but this time with respect to its underlying effect for a kernel size of KWxKH. - \f$\lambda\f$ wave length of pixels (e.g. if set to KW, the wave will oscillate exactly once for direction (\f$\theta=0\f$) - \f$\theta\f$ orientation of the wave (0-> wave direction is 3 o'clock, \f$\frac{\pi}{2}\f$ -> wave direction is 12 o'clock) - \f$\psi\f$ phase shift of the wave - \f$\sigma\f$ std.deviation of the Gaussian multiplied with the wave (in pixels) - \f$\gamma\f$ aspect-ratio of the Gaussian **/ class ICLFilter_API GaborOp : public UnaryOp, public utils::Uncopyable{ public: /// creates an empty GaborOp GaborOp(); /// creates a new gabor op with given kernel size and parameters /** The Gabor-Jet internally created consist of one convolution kernel for each possible combination of the parameter values. E.g. if the parameters are: - lambdas = {5} - thetas = {1,2} - psis = {0} - sigmas = {100,200} - and gammas = {1] the gabor jet consist of 4 convolution kernels with fixed params \f$\lambda=5\f$, \f$\psi=0\f$, \f$\gamma=1\f$ and variable params -# \f$\theta=1\f$ and \f$\sigma=100\f$ -# \f$\theta=2\f$ and \f$\sigma=100\f$ -# \f$\theta=1\f$ and \f$\sigma=200\f$ -# \f$\theta=2\f$ and \f$\sigma=200\f$ */ GaborOp(const utils::Size &kernelSize, std::vector lambdas, std::vector thetas, std::vector psis, std::vector sigmas, std::vector gammas ); ~GaborOp(); /// sets the current kernel size /** if the kernels have already been created, they are updated to this new size value */ void setKernelSize(const utils::Size &size); /// add a new lambda value void addLambda(float lambda); /// add a new theta value void addTheta(float theta); /// add a new psi value void addPsi(float psi); /// add a new sigma value void addSigma(float sigma); /// add a new gamma value void addGamma(float gamma); /// update the current kernels by currently possible value combinations void updateKernels(); /// apply all filters to an image /** The output image gets as many channels as kernels could be created by combining given parameters. Channels c of ppoDst is complies the convolution result of the c-th kernel. */ virtual void apply(const core::ImgBase *poSrc, core::ImgBase **ppoDst); /// Import unaryOps apply function without destination image using UnaryOp::apply; /// apply all filters to an image at a specific position /** The result vector contains the filter-response for all kernels */ std::vector apply(const core::ImgBase *poSrc, const utils::Point &p); /// static function to create a gabor kernel by given gabor parameters /** As reminder: \f[ g(x,y;\lambda,\theta,\psi,\sigma,\gamma)=\exp(-\frac{x'^2+\gamma^2y'^2}{2\sigma^2})\cos(2\pi\frac{x'}{\lambda}+\psi) \f] where \f[ x' = x \cos\theta + y \sin\theta\, \f] and \f[ y' = -x \sin\theta + y \cos\theta\, \f] */ static core::Img32f *createKernel(const utils::Size &size, float lambda, float theta, float psi, float sigma, float gamma); /// returns all currently created kernels const std::vector &getKernels() const { return m_vecKernels; } private: std::vector m_vecLambdas; std::vector m_vecThetas; std::vector m_vecPsis; std::vector m_vecSigmas; std::vector m_vecGammas; std::vector m_vecKernels; std::vector m_vecResults; utils::Size m_oKernelSize; }; } // namespace filter }